Geophys. J. Int.

(1999) 139, 806^822

Introduction to the spectral element method for three-dimensional

seismic wave propagation

Dimitri Komatitsch and Jeroen Tromp

Department of Earth and Planetary Sciences, Harvard University, Cambridge, MA 02138, USA. E-mail: komatits@seismology.harvard.edu

Accepted 1999 July 14. Received 1999 July 12; in original form 1999 March 29

Downloaded from https://academic.oup.com/gji/article/139/3/806/587264 by Indian Institute of Technology Madras user on 08 November 2023
S U M M A RY
We present an introduction to the spectral element method, which provides an
innovative numerical approach to the calculation of synthetic seismograms in 3-D
earth models. The method combines the £exibility of a ¢nite element method with the
accuracy of a spectral method. One uses a weak formulation of the equations of motion,
which are solved on a mesh of hexahedral elements that is adapted to the free surface
and to the main internal discontinuities of the model. The wave¢eld on the elements is
discretized using high-degree Lagrange interpolants, and integration over an element
is accomplished based upon the Gauss^Lobatto^Legendre integration rule. This com-
bination of discretization and integration results in a diagonal mass matrix, which
greatly simpli¢es the algorithm. We illustrate the great potential of the method by
comparing it to a discrete wavenumber/re£ectivity method for layer-cake models. Both
body and surface waves are accurately represented, and the method can handle point
force as well as moment tensor sources. For a model with very steep surface topography
we successfully benchmark the method against an approximate boundary technique.
For a homogeneous medium with strong attenuation we obtain excellent agreement
with the analytical solution for a point force.
Key words: attenuation, ¢nite element methods, numerical techniques, seismic
modelling, seismic wave propagation, topography.

homogeneous regions. In three dimensions, the numerical

1 IN T ROD U C T I O N
cost is high and approximations need to be made that lead to
In both regional and global seismology, the accurate calculation artefacts in the solution (Bouchon et al. 1996). Classical ¢nite
of seismograms in realistic 3-D earth models has become a element methods have been successfully applied to the study of
necessity. A large collection of numerical techniques is available wave propagation in 3-D sedimentary basins (Bao et al. 1998).
for this purpose. Among them, the most widely used approach These techniques surmount some of the previously mentioned
is probably the ¢nite di¡erence method (e.g. Kelly et al. 1976; di¤culties, but come with a high computational cost due
Virieux 1986). This approach has been used to calculate the to the fact that large linear systems need to be solved. The
wave¢eld in 3-D local and regional models (e.g. Olsen & implementation of such algorithms on parallel computers with
Archuleta 1996; Graves 1996; Ohminato & Chouet 1997). distributed memory complicates matters further (Bao et al.
Unfortunately, signi¢cant di¤culties arise in the presence of 1998). A promising new approach that combines aspects of FD,
surface topography (Robertsson 1996) and when anisotropy ¢nite element and discrete wavenumber modelling has been
needs to be incorporated (Igel et al. 1995). Pseudospectral proposed to reduce signi¢cantly the cost of the simulations
methods have become popular for regional (Carcione 1994; (Moczo et al. 1997). Another approach is the direct solution
Tessmer & Koslo¡ 1994) and global (Tessmer et al. 1992; method developed by Geller & Ohminato (1994) speci¢cally
Furumura et al. 1998) problems, but are restricted to models for problems in global seismology. As usual in a Galerkin
with smooth variations. Because of the problems associated with method, it involves the manipulation of large matrices and an
the implementation of the free-surface boundary condition, approximate treatment of boundary undulations.
the accurate representation of surface waves in both ¢nite The spectral element method discussed in this article has
di¡erence (FD) and pseudospectral methods is a di¤cult been used for more than 15 years in computational £uid
problem and an active area of research (Robertsson 1996; dynamics (Patera 1984). It has recently gained interest for
Graves 1996; Komatitsch et al. 1996). Boundary integral problems related to 2-D (Seriani et al. 1992; Cohen et al. 1993;
methods provide an elegant approach for incorporating topo- Priolo et al. 1994) and 3-D (Komatitsch 1997; Faccioli et al.
graphic variations, but are restricted to a ¢nite number of 1997; Komatitsch & Vilotte 1998; Seriani 1998; Komatitsch

806 ß 1999 RAS

 Seismic spectral element method 807

et al. 1999c; Paolucci et al. 1999) seismic wave propagation. increasing xi are denoted by xª i, and partial derivatives with
The method easily incorporates free-surface topography and respect to xi are denoted by Li.
accurately represents the propagation of surface waves. The The equations of motion that govern the propagation of
e¡ects of anisotropy (Seriani et al. 1995; Komatitsch et al. seismic waves in an elastic or anelastic solid may be solved
1999b) and £uid^solid boundaries (Komatitsch et al. 1999a) based upon either a strong or a weak formulation of the
can also be accommodated. The method lends itself well to problem. In the strong formulation one works directly with
parallel computation with distributed memory (Fischer & the equations of motion and associated boundary conditions
RÖnquist 1994; Seriani 1997; Komatitsch & Vilotte 1998). written in di¡erential form; this approach is used, for instance,
The purpose of this article is to give a detailed introduction in FD or global pseudospectral modelling techniques. In the
to the spectral element method in the context of seismic wave weak formulation one uses an integral form of the equations
propagation. We demonstrate the accuracy of the method for of motion, as in ¢nite element (FEM) and direct solution
both body and surface waves by comparing its results against methods. The spectral element method (SEM) discussed in

Downloaded from https://academic.oup.com/gji/article/139/3/806/587264 by Indian Institute of Technology Madras user on 08 November 2023
those of a discrete wavenumber/re£ectivity method (Bouchon this article is based upon a weak formulation of the equations
1981; Mu«ller 1985) for simple layer-cake models. Simulations of motion. We shall ignore the e¡ects of £uid regions and
for a full moment tensor source are compared against results associated £uid^solid boundaries, although the SEM can be
based upon a frequency^wavenumber and an FD method used to solve such problems (Komatitsch et al. 1999a).
(Graves 1996) for a dip-slip source in a half-space. We also
demonstrate that the method can handle steep topography by
analysing the e¡ects of a hemispherical crater embedded in 2.1 Strong form
a half-space on an incident plane compressional wave, and The displacement ¢eld s produced by an earthquake is governed
comparing the results to those obtained with an approximate by the momentum equation,
boundary method (Sänchez-Sesma 1983). Finally, we show that
strong attenuation can be taken into account by simulating a o L2t s~= . Tzf . (1)
2-D homogeneous medium with QP ^30 and QS ^20 and The distribution of density is denoted by o. The stress tensor T
comparing the results to the analytical solution for a point is linearly related to the displacement gradient =s by Hooke's
source force derived by Carcione et al. (1988). law, which in an elastic, anisotropic solid may be written in the
form
T~c5=s . (2)
2 E QUAT IO N S OF MOT I O N
The elastic properties of the earth model are determined by
We seek to determine the displacement ¢eld produced by an the fourth-order elastic tensor c, which has 21 independent
earthquake in a ¢nite earth model with volume ), as shown in components in the case of general anisotropy. Modelling
Fig. 1. The boundaries of this volume include a stress-free wave propagation in fully anisotropic media with classical
surface L), as well as an absorbing boundary !. Seismic waves techniques such as the FD method is di¤cult because of the use
are re£ected by the free surface L); ideally, they are completely of staggered grids (Igel et al. 1995). The SEM, on the other
absorbed by the arti¢cial boundary !. The unit outward hand, can accurately model the e¡ects of any anisotropy on
normal to the boundary L)z! is denoted by nª. The earth elastic waves (Komatitsch et al. 1999b).
model may have any number of internal discontinuities; the In an attenuating medium, Hooke's law (2) needs to be
unit upward normal to such discontinuities is also denoted by nª. modi¢ed such that the stress is determined by the entire strain
Locations within the model are denoted by the position vector history:
x~(x, y, z). For brevity, a component of the position vector will ?
sometimes be denoted using index notation: xi , i~1, 2, 3, T(t)~ Lt c(t{t0 )5=s(t0 ) dt0 . (3)
where x1 ~x, x2 ~y, and x3 ~z. Unit vectors in the directions of {?

In seismology, the quality factor Q is observed to be constant

over a wide range of frequencies. Such an absorption-band solid
may be mimicked by a series of L standard linear solids (Liu
et al. 1976). In practice, two or three linear solids usually su¤ce
to obtain an almost constant Q (Emmerich & Korn 1987). The
components of the anelastic tensor for such a series may be
written in the form
" #
XL
R ` p` {t=qp`
cijkl (t)~cijkl 1{ (1{qijkl /q ) e H(t) , (4)
`~1

where cRijkl denotes the relaxed modulus and H(t) is the

Heaviside function. The stress relaxation times qp` are chosen
to be the same for all components of the anelastic tensor,
Figure 1. Finite earth model with volume ) and free surface L). An
whereas the strain relaxation times q`ijkl are unique to each
arti¢cial absorbing boundary ! is introduced, and nª denotes the unit individual component. In practice, attenuation in the earth is
outward normal to all boundaries. The model can be fully hetero- mainly controlled by the shear quality factor, such that only
geneous or composed of any number of layers. The source xs can be the time dependence of the isotropic shear modulus need be
placed anywhere inside ). accommodated.

ß 1999 RAS, GJI 139, 806^822

808 D. Komatitsch and J. Tromp

Using the absorption-band anelastic tensor (4), the

2.2 Weak form
constitutive relation (3) may be rewritten in the form
X
L Rather than using the equations of motion and associated
T~cU 5=s{ R` , (5) boundary conditions directly, one can use an integrated form.
`~1 This is accomplished by dotting the momentum equation (1)
with an arbitrary test vector w, integrating by parts over
where for each standard linear solid
the model volume ), and imposing the stress-free boundary
Lt R` ~{R` /qp` zäc` 5=s/qp` . (6) condition (10) and the absorbing boundary condition expressed
in terms of traction (11). This gives
The components of the unrelaxed modulus cU
ijkl are given by
" #
U R
X
L
` p`
o w . L2t s d 3 x
cijkl ~cijkl 1{ (1{qijkl /q ) , (7) )

Downloaded from https://academic.oup.com/gji/article/139/3/806/587264 by Indian Institute of Technology Madras user on 08 November 2023
`~1

and the modulus defect äc` associated with each individual ~{ =w5T d 3 xzM5=w(xs ) S(t)
)
standard linear solid is determined by
dc`ijkl ~{cR ` p`
ijkl (1{qijkl /q ) . (8) z o[on (nª . Lt s)nª zo1 ( tê 1 . Lt s) tê 1 zo2 ( tê 2 . Lt s) tê 2 ] . w d 2 x ,
!
The earthquake source is represented by the point force f,
(12)
which may be written in terms of a moment tensor M as
f~{M . =d(x{xs )S(t) . (9) where the stress tensor T is determined in terms of the dis-
placement
„ gradient =s by Hooke's law (2). The source term,
The location of the point source is denoted by xs, d(x{xs ) . 3
) f w d x, has been explicitly integrated using the properties
denotes the Dirac delta distribution located at xs , and the of the Dirac delta distribution. Slip on a ¢nite fault plane may
source time function is given by S(t). A ¢nite source may be be simulated by making the substitution
simulated by using a number of point sources, each with its
own time history. M5=w(xs ) S(t)? m(xs , t)5=w(xs ) d 2 xs , (13)
The momentum equation (1) must be solved subject to a Ss
stress-free boundary condition at the earth's surface L): where Ss denotes the fault surface and m(xs , t) denotes the
T . nª ~0 . (10) moment density tensor.
Mathematically, the strong and the weak formulations are
In FD and global pseudospectral applications, the imple- equivalent because (12) holds for any test vector w. However,
mentation of the free-surface condition (10), especially in the one of the nice features of the weak formulation is that the
presence of surface topography, is a di¤cult problem and an stress-free surface boundary condition (10) is naturally satis-
active area of research (e.g. Carcione 1994; Tessmer & Koslo¡ ¢ed; that is, it does not have to be imposed explicitly. The
1994; Komatitsch et al. 1996; Moczo et al. 1997; Ohminato & incorporation of free-surface topography is therefore straight-
Chouet 1997). forward, and surface waves are more accurately simulated than
At every internal boundary, both the displacement s and in methods based upon the strong form (Komatitsch & Vilotte
the traction T . nª need to be continuous. At the arti¢cial model 1998; Komatitsch et al. 1999c).
boundary !, waves travelling out of the volume ) need to be The weak form of the equations of motion (12) remains
absorbed. The implementation of an e¤cient absorbing valid in an anelastic earth model, except that the stress tensor T
boundary is not an easy problem (Clayton & Engquist 1977; is in this case determined by the absorption-band constitutive
Quarteroni et al. 1998; Komatitsch et al. 1999c). The approxi- relation (5), and that memory in the system is accounted for by
mate absorbing boundary condition we shall use relates eq. (6). This approach has been used in the context of ¢nite
traction to velocity, element simulations by Moczo et al. (1997) and Kay & Krebes
T . nª ~o[on (nª . Lt s)nª zo1 ( tê 1 . Lt s) tê 1 zo2 ( tê 2 . Lt s) tê 2 ] , (11) (1999), and is more satisfactory than the use of a simple
damping matrix (Bao et al. 1998; Faccioli et al. 1997), which
where tê 1 and tê 2 are orthogonal unit vectors tangential to the results in a Q that strongly depends on frequency. The main
absorbing boundary ! with unit outward normal nª , on is the drawback of the approach, however, is the additional computer
quasi-P wave speed of waves travelling in the nª direction, o1 is memory requirement resulting from the use of several memory
the quasi-S wave speed of waves polarized in the tê 1 direction, variables per gridpoint. Spreading the memory variables over
and o2 is the quasi-S wave speed of waves polarized in the tê 2 neighbouring gridpoints has been suggested to reduce this
direction. The absorbing boundary condition (11) is based upon requirement (Zeng 1996; Day 1998).
a one-way treatment that perfectly absorbs waves impinging at
right angles to the boundary, but that is less e¡ective for waves
that graze the boundary (Clayton & Engquist 1977). It is valid
3 D E F I N I T I O N OF T H E M E S H
for transversely isotropic media with a horizontal or vertical
symmetry axis; more general anisotropy can be accommodated As in a classical FEM, the model volume ) is subdivided into
by tapering it such that the medium becomes transversely iso- a number of non-overlapping elements )e , e~1, . . . , ne , such
tropic on the absorbing boundary !. Note that one can use that )~|ne e )e , as shown in Fig. 2. As a result of this sub-
attenuation close to the arti¢cial absorbing boundary of the division, the absorbing boundary ! is similarly represented
model to increase the e¤ciency of the approximate absorbing by a number of surface elements !b , b~1, . . . , nb , such that
condition (11). !~|nbb !b . In a FEM, a variety of elements such as tetrahedra

ß 1999 RAS, GJI 139, 806^822

 Seismic spectral element method 809

square " may therefore be written in the form

X
na
x(m, g)~ Na (m, g)xa . (14)
a~1

The shape functions Na (m, g) are products of Lagrange poly-

nomials of degrees 1 or 2. The n` z1 Lagrange polynomials
of degree n` are de¢ned in terms of n` z1 control points
{1¦ma ¦1, a~0, . . . , n` , by
(m{m0 )


(m{ma{1 )(m{maz1 )


(m{mn` )
`na` (m)~ . (15)
(ma {m0 )


(ma {ma{1 )(ma {maz1 )


(ma {mn` )

Downloaded from https://academic.oup.com/gji/article/139/3/806/587264 by Indian Institute of Technology Madras user on 08 November 2023
Notice from this de¢nition that when the Lagrange poly-
Figure 2. For the purpose of computations, the earth model )
shown in Fig. 1 is subdivided into curved elements (quadrangles in 2-D,
nomials are evaluated at a control point mb they return a value
hexahedra in 3-D) whose shapes are adapted to the edges of the model of either 0 or 1:
L) and !, as well as to the main geological interfaces. `na` (mb )~dab , (16)
where d denotes the Kronecker delta. The two Lagrange
or hexahedra can be used (e.g. Dhatt & Touzot 1984), but polynomials of degree 1 with two control points, m~{1 and
a SEM is restricted to hexahedral volume elements )e . m~1, are `10 (m)~(1{m)/2 and `11 (m)~(1zm)/2, and the three
Consequently, the absorbing boundary ! is subdivided in Lagrange polynomials of degree 2 with three control points,
terms of quadrilateral surface elements !b . m~{1, m~0 and m~1, are `20 (m)~m(m{1)/2, `21 (m)~1{m2 and
In the next two sections we brie£y describe how boundary `22 (m)~m(mz1)/2. As an example, the four shape functions
and volume elements are parametrized. In what follows the associated with the four-anchor quadrilateral element shown in
reader should keep in mind that by `boundary elements' we Fig. 3 are products of degree 1 Lagrange polynomials:
simply mean the sides of some of the volume elements. The N1 (m, g)~`10 (m)`10 (g), N2 (m, g)~`11 (m)`10 (g), N3 (m, g)~`11 (m)`11 (g)
results in this section may be found in any ¢nite element text and N4 (m, g)~`10 (m)`11 (g). Similarly, the shape functions of
book (Zienkiewicz 1977; Hughes 1987). We include them here nine-anchor quadrilateral elements are products of degree 2
in an e¡ort to provide a more complete overview of the SEM. Lagrange polynomials.
The orientation of the reference square " is chosen such that
the unit outward normal nª to boundary element !b is given by
3.1 Boundary elements 1 Lx Lx
nª ~ | , (17)
Each quadrilateral boundary element is isomorphous to the Jb Lm Lg
square, hence there exists a unique mapping from the square to where Jb denotes the Jacobian of the transformation
each surface element !b , b~1, . . . , nb . The reference square
is de¢ned in terms of greek coordinates (m, g), {1¦m¦1, Lx Lx
Jb ~ | . (18)
{1¦g¦1, which are sometimes referred to as the natural Lm Lg
coordinates. Each quadrilateral element !b is de¢ned in terms The boundary elements should be de¢ned in such a way that
of a set of na control points or anchors xa ~x(ma , ga ), the Jacobian Jb never vanishes. As in a classical FEM, this
a~1, . . . , na , and a set of na shape functions Na (m, g). Its four places important constraints on the mesh generation process
corners are always used as anchors, but its side centres and its (Hughes 1987), which are further discussed later. To calculate
centre may be used as additional anchors. For simple boundary the Jacobian Jb and the unit outward normal nª one needs to
elements with straight edges, four control points su¤ce, determine the six partial derivatives Lx/Lm and Lx/Lg. This is
whereas for curved boundary elements nine control points may accomplished by di¡erentiating the mapping (14):
be needed to describe their shape accurately. Fig. 3 illustrates
quadrilateral elements de¢ned in terms of four and nine control X
na
Lm x(m, g)~ Lm Na (m, g)xa ,
points. For any given boundary element, the relation between a a~1
point x within the element !b and a point (m, g) in the reference (19)
X
na
Lg x(m, g)~ Lg Na (m, g)xa .
a~1

Partial derivatives of the shape functions, Lm Na and Lg Na , are

analytically determined in terms of Lagrange polynomials of
degrees 1 or 2 and their derivatives.

3.2 Volume elements

Similarly, each hexahedral volume element )e can be mapped
to a reference cube. Points within this reference cube are
Figure 3. The geometry of each of the curved surface quadrangles can denoted by the vector î~(m, g, f), where {1¦m¦1, {1¦g¦1
be de¢ned by either four control nodes (left) or nine control nodes and {1¦f¦1. The components of the vector î will sometimes
(right). be denoted using index notation: mi , i~1, 2, 3, where m1 ~m,

ß 1999 RAS, GJI 139, 806^822

810 D. Komatitsch and J. Tromp

m2 ~g, m3 ~f. Analogous to the boundary elements discussed

4 R E PR E S E N TAT I O N OF F U NC T I O N S O N
in the previous section, each volume element is de¢ned in
T HE E L E M E N T S
terms of na anchors xa ~x(ma , ga , fa ), a~1, . . . , na , and shape
functions Na (î). At least eight corner nodes are needed to To solve the weak form of the equations of motion (12),
de¢ne a hexahedral volume element; by adding mid-side and integrations over the volume ) and the absorbing boundary !
centre nodes the number of anchors can become as large as 27. are subdivided in terms of smaller integrals over the volume
Fig. 4 illustrates eight-node and 27-node hexahedral elements. and surface elements )e and !b , respectively. This section is
The mapping between points within the hexahedral element concerned with the representation of functions on the elements,
and the reference cube may be written in the form and with the integration of functions over an element.
We have seen in the previous section that the shape of the
X
na
boundary and volume elements can be de¢ned in terms of
x(î)~ Na (î)xa . (20)
low-degree Lagrange polynomials. In a traditional FEM, low-

Downloaded from https://academic.oup.com/gji/article/139/3/806/587264 by Indian Institute of Technology Madras user on 08 November 2023
a~1
degree polynomials are also used as basis functions for the
Shape functions of eight-node hexahedral elements are triple representation of ¢elds on the elements. In a SEM, on the other
products of degree 1 Lagrange polynomials, and those of hand, a higher-degree Lagrange interpolant is used to express
27-node hexahedral elements are triple products of degree 2 functions on the elements. Of course, one could use the same
Lagrange polynomials. Frequently one uses a reduced 20-node high-degree polynomial representation to de¢ne the geo-
element (Dhatt & Touzot 1984), which is obtained by eliminating metrical mapping on each element, but in practice such high
seven centre nodes from the 27-node element, as illustrated in precision is unnecessary, provided the variations in element
Fig. 4. geometry are smooth enough.
An element of volume dx dy dz within a given element )e For reasons discussed in Sections 5 and 6, the control points
is related to an element of volume dm dg df in the reference ma , a~0, . . . , n` , needed in the de¢nition (15) of the Lagrange
cube by polynomials of degree n` are chosen to be the n` z1 Gauss^
Lobatto^Legendre points, which are the roots of
dx dy dz~Je dm dg df , (21)
(1{m2 )Pn0 ` (m)~0 , (24)
where the volumetric Jacobian Je is given by

L(x, y, z) where Pn0 ` denotes the derivative of the Legendre polynomial

Je ~ . (22) of degree n` . These points can be computed by numerical
L(m, g, f)
resolution of (24) (Canuto et al. 1988, p. 61). Note that the
To calculate the Jacobian Je , we need the partial derivative Gauss^Lobatto^Legendre points always include z1 and {1;
matrix Lx/Lî, which is obtained by di¡erentiating the therefore in a SEM some points always lie exactly on the
mapping (20): boundaries of the elements. As an example, Fig. 5 illustrates
the nine Lagrange polynomials of degree 8 based upon the nine
Lx Xna
LNa Gauss^Lobatto^Legendre control points de¢ned by (24).
~ xa . (23)
Lî a~1 Lî

Partial derivatives of the shape functions are determined 4.1 Polynomial representation on elements
analytically in terms of Lagrange polynomials of degree 1 or 2
and their derivatives. The elements should be constructed in On boundary elements !b , a function f is interpolated by
such a way that the Jacobian Je never vanishes, which again products of Lagrange polynomials of degree n` as
poses strong constraints on the mesh generation process. This
X
n`
ensures that the mapping from the reference cube to the f (x(m, g))& f ab `a (m)`b (g) . (25)
element, x(î), is unique and invertible, that is, î(x) is well a,b~0
de¢ned.
For convenience, the polynomial degree n` has been omitted as
a superscript on the Lagrange polynomials. As a result of the
de¢nition (16) of the Lagrange polynomials, the coe¤cients
f ab are the functional values of f at the interpolation points
x(ma , gb ):

f ab ~f (x(ma , gb )) . (26)

In a SEM for wave propagation problems one typically uses a

polynomial degree n` between 5 and 10 to represent a function
on the element (Seriani & Priolo 1994; Komatitsch & Vilotte
Figure 4. The geometry of each of the curved volume hexahedra can 1998). Similarly, on each volume element )e a function f is
be de¢ned by either eight control nodes (left) or 27 control nodes interpolated by triple products of Lagrange polynomials of
(right). In the case of the 27-node brick, the empty squares indicate degree n` as
the six nodes that lie in the middle of the sides of the element, and the
triangle indicates the node that lies at the centre of the element. By X
n`
ignoring these nodes, one obtains the classical 20-node element that f (x(m, g, f))& f abc `a (m)`b (g)`c (f),
can also be used to de¢ne the geometry. a,b,c~0

ß 1999 RAS, GJI 139, 806^822

 Seismic spectral element method 811

Downloaded from https://academic.oup.com/gji/article/139/3/806/587264 by Indian Institute of Technology Madras user on 08 November 2023
Figure 5. Lagrange interpolants of degree N~8 at the Gauss^Lobatto^Legendre points on the reference segment [{1, 1]. The Nz1~9
Gauss^Lobatto^Legendre points can be distinguished along the horizontal axis. All Lagrange polynomials are, by de¢nition, equal to 1 or 0 at each
of these points.

where rule,
1 1
f abc ~f (x(ma , gb , fc )) . (27)
f (x) d 2 x~ f (x(m, g)) Jb (m, g) dm dg
!b {1 {1
Using this polynomial representation, the gradient of a
function, =f , may be written in the form X
n`
& ua ub f ab Jbab , (30)
a,b~0
X
3
=f (x(m, g, f))& xª i Li f (x(m, g, f))
where ua , a~0, . . . , n` , denote the weights associated
i~1
with the Gauss^Lobatto^Legendre points of integration, and
X
3 X
n` Jbab ~Jb (ma , gb ). These weights are computed numerically
~ xª i f abc [`0a (m)`b (g)`c (f)Li m (Canuto et al. 1988, p. 61).
i~1 a,b,c~0 Similarly, integrations over volume elements )e may be
approximated as
z`a (m)`0b (g†`c f)Li gz`a (m)`b (g)`0c (f)Li f] , (28)
1 1 1
where a prime denotes di¡erentiation. When evaluated at f (x) d 3 x~ f (x(m, g, f)) Je (m, g, f) dm dg df
)e {1 {1 {1
any of the Gauss^Lobatto^Legendre points x(ma0 , gb0 , fc0 ), this
expression reduces to X
n`
& ua ub uc f abc Jeabc , (31)
X
3 X
n` X
n` a,b,c~0
0 0 0 0
=f (x(m, g, f))& xª i f ab c `0a (ma0 )Li mz f a bc `0b (gb0 )Li g
i~1 a~0 b~0 where Jeabc ~Je (ma , gb , fc ). To facilitate the integration of
 functions and their partial derivatives over the elements, the
X
n`
z
0 0
f a b c `0c (fc0 )Li f . (29) values of the inverse Jacobian matrix Lî/Lx need to be deter-
c~0 mined at the (n` z1)3 Gauss^Lobatto^Legendre integration
points for each element.
Notice that in (29), which is based upon the chain rule,
one needs the inverse Jacobian matrix Lî/Lx. This matrix
5 G L O BA L S YST EM A N D T I M E
is obtained by inverting the Jacobian matrix Lx/Lî, which is
MA RC H I NG
determined in terms of the shape functions by (23). This inverse
exists provided the Jacobian Je is non-singular, which is a We have seen that in a SEM the model is subdivided in terms of
requirement for the proper design of the mesh, as mentioned a number of hexahedral elements. On each individual element,
previously. functions are sampled at the Gauss^Lobatto^Legendre points
of integration. Gridpoints that lie on the sides, edges or corners
of an element are shared amongst neighbouring elements, as
4.2 Integration over elements
illustrated in Fig. 6. Therefore, the need arises to distinguish
At this stage, integrations over surface elements !b may be between the gridpoints that de¢ne an element, the local mesh,
approximated using the Gauss^Lobatto^Legendre integration and all the gridpoints in the model, many of which are shared

ß 1999 RAS, GJI 139, 806^822

812 D. Komatitsch and J. Tromp

construction of the global mass and sti¡ness matrices from

their elemental forms may be found in Komatitsch & Vilotte
(1998).
A highly desirable property of a SEM, which allows for a
very signi¢cant reduction in the complexity and cost of the
algorithm, is the fact that the mass matrix M is diagonal by
construction. Therefore, no costly linear system resolution
algorithm is needed to march the system in time. At the
elemental level, the mass matrix is given by the ¢rst term in
the weak formulation of the equations of motion (12):

o w . L2t s d 3 x

Downloaded from https://academic.oup.com/gji/article/139/3/806/587264 by Indian Institute of Technology Madras user on 08 November 2023
)e

1 1 1
~ o(x(î)) w(x(î)) . L2t s(x(î), t) J(î) d 3 î . (33)
{1 {1 {1

The displacement s is expanded on an element as

X
3 X
n`
s(x(m, g, f), t)& xª j spql
j (t)`p (m)`q (g)`l (f) . (34)
j~1 p,q,l~0

Similarly, we choose test functions of the form

Figure 6. Illustration of the local and global meshes for a four- X

3 X
n`
w(x(m, g, f))~ xª i wabc
i `a (m)`b (g)`c (f) . (35)
element 2-D spectral element discretization with polynomial degree
i~1 a,b,c~0
N~4. Each spectral element contains (Nz1)2 ~25 Gauss^Lobatto^
Legendre points, which constitute the local mesh for each element. Upon substituting (35) and (34) into (33), and introducing the
These points are non-evenly spaced, but have been drawn evenly Gauss^Lobatto^Legendre numerical integration, we ¢nd that
spaced here for simplicity. In the global mesh, points lying on edges or
corners (as well as on faces in three dimensions) are shared between
o w . L2t s d 3 x
elements. The contributions to the global system of degrees of freedom, )e
computed separately on each element, have to be summed at these
common points represented by black dots. Exactly two elements share X 0 0 0 0 0 0
X
3

points inside an edge in two dimensions, while corners can be shared by & ua0 ub0 uc0 Jea b c oa b c xª i . xª j
any number of elements depending on the topology of the mesh, which a0 ,b0 ,c0 i, j~1

can be non-structured. X X
| wabc
i `a (ma0 )`b (gb0 )`c (fc0 ) s pql
j (t)`p (ma0 )`q (gb0 )`l (fc0 )
a,b,c p,q,l

X 0 0 0 0 0 0
X
3 X
amongst several spectral elements, the global mesh. One needs ~ ua0 ub0 uc0 Jea b c oa b c dij wabc
i daa0 dbb0 dcc0
to determine a mapping between gridpoints in the local mesh a0 ,b0 ,c0 i, j~1 a,b,c
and gridpoints in the global mesh; e¤cient routines are avail- X
able for this purpose from ¢nite element modelling. Before the | spql
i (t)dpa0 dqb0 dlc0
p,q,l
system can be marched forward in time, the contributions from
all the elements that share a common global gridpoint need X X
3

to be summed. In a traditional FEM this is referred to as ~ ua ub uc Jeabc oabc wabc

i s abc
i (t) , (36)
a,b,c i~1
the assembly of the system. Computationally, this assembly
stage is a costly part of the calculation on parallel computers where oabc ~o(x(ma , gb , fc )), and where a dot denotes di¡er-
because information from individual elements needs to be entiation with respect to time. It is important to realize
shared with neighbouring elements, an operation that involves that the weak form of the equations of motion holds for
communication between distinct CPUs. any test vector w. Therefore, the global system (32) is built
Let U denote the displacement vector of the global system; by independently setting factors of wabc 1 , w2
abc
and wabc
3 equal
that is, U contains the displacement vector at all the gridpoints to zero. What is remarkable about the result (36) is the fact
in the global mesh, classically referred to as the global degrees that the value of acceleration at each point, s abc i (t), is simply
of freedom of the system. The ordinary di¡erential equation multiplied by the factor ua ub uc oabc Jeabc ; that is, the elemental
that governs the time dependence of the global system may be mass matrix is diagonal. This property also holds true for
written in the form the global mass matrix after assembly of the system. It is this
 _ desirable property that has motivated the use of Lagrange
M UzC UzKU~F , (32)
interpolants for the representation of functions on the elements,
where M denotes the global mass matrix, C the global in conjunction with the use of the Gauss^Lobatto^Legendre
absorbing boundary matrix, K the global sti¡ness matrix integration rule. Note that density o need not be constant over
and F the source term. Explicit expressions for the local con- an element, but may vary from one gridpoint to another.
tributions to the sti¡ness matrix, the source and the absorbing Therefore, the SEM is able to handle fully heterogeneous
boundaries are given in the Appendix. Further details on the media.

ß 1999 RAS, GJI 139, 806^822

 Seismic spectral element method 813

In this study, time discretization of the second-order ordinary practice, one generally designs the mesh such that it honours
di¡erential equation (32) is achieved based upon a classical the free surface and the main geological boundaries within the
explicit second-order FD scheme, moving the sti¡ness and model. As in a traditional FEM, the curvature of the elements
absorbing terms to the right-hand side. Such a scheme is con- should be modest, such that the Jacobian of the mapping to the
ditionally stable, and the Courant stability condition is reference domain varies smoothly across any given element
governed by the minimum value of the ratio between the size and never vanishes (Hughes 1987).
of the grid cells and the P-wave velocity. In some cases, for Because seismic wave speed generally increases with depth,
instance when very small cells are needed in high wave-speed one often needs to coarsen the grid in the deeper parts of the
regions, the Courant stability condition may be too restrictive model to retain a similar number of gridpoints per wavelength.
for an explicit scheme. In that case a Newmark scheme, written A simple way of accomplishing this in two stages consists
in prediction^multicorrection format, can be used instead of doubling the grid ¢rst in one horizontal dimension and,
(Komatitsch 1997; Komatitsch et al. 1999a). For anelastic subsequently, at a greater depth, in the second horizontal

Downloaded from https://academic.oup.com/gji/article/139/3/806/587264 by Indian Institute of Technology Madras user on 08 November 2023
media, we separately march the strong form of the memory dimension, as illustrated in Fig. 7. This mesh design will prove
variable eq. (6) in time, regarding the term in =s as a source, useful for studying a layer-cake model with low wave speeds
as proposed by Robertsson et al. (1994). These ¢rst-order near the surface in Section 7.1. However, this simple con-
equations in time are integrated based upon a fourth-order forming mesh is not optimal because it contains 450 angles, and
Runge^Kutta scheme that is known to be e¤cient for this therefore high local variations in the size and shape of the
problem (Carcione 1994). spectral elements. In this respect, an area of active research
involves the use of non-conforming meshes in which two or
more elements share the same edge with another, larger
6 AC C U R ACY OF T H E MET HOD
element, or the combination of spectral element algorithms
with other techniques such as classical FEMs (Bernardi et al.
6.1 Selection of the polynomial degree
1990; Lahaye et al. 1997; Chaljub & Vilotte 1998). These
In a SEM, spatial resolution is controlled by two parameters: approaches simplify the coarsening of a mesh considerably, but
the typical size of an element, *h, and the polynomial degree are signi¢cantly more di¤cult to implement.
used to represent functions on an element, n` , each element
thus containing n` z1 points in each direction. In this respect,
6.3 Accuracy of the integration
SEMs are related to FEMs based upon a high polynomial
degree, the so-called h-p version of FEM (Guo & Babus̄ka We have seen in Section 5 that the use of Lagrange inter-
1986). In practice, if the polynomial degree n` is too small, e.g. polants for the representation of functions on an element,
less than typically 4, a SEM exhibits the same inaccuracies that in conjunction with the use of a Gauss^Lobatto^Legendre
are observed in a standard FEM applied to wave propagation numerical integration, results in a diagonal mass matrix, which
problems (Marfurt 1984). On the other hand, if the polynomial in turn leads to a simple time integration scheme. This has
degree is very large, e.g. greater than 15, the method is spatially been accomplished based upon the Gauss^Lobatto^Legendre
very accurate, but the computational requirements become integration rule, rather than the more classical Gauss rule that
prohibitive. The reason for this is that the evaluation of the
sti¡ness matrix at the elemental level has a cost of O(n4` ) in
three dimensions due to the matrix multiplications involved.
Typically, polynomial degrees between 5 and 10 are therefore
optimal for a SEM applied to wave propagation problems,
because this provides the best trade-o¡ between accuracy and
cost (Seriani & Priolo 1994). In order to obtain accurate
results, *h has to be chosen such that the average number
of points per minimum wavelength jmin in an element,
jmin ~(n` z1)/*h, is roughly equal to 5 for the optimal range of
polynomial degrees mentioned above (Seriani & Priolo 1994;
Faccioli et al. 1997). Therefore, a SEM compares favourably
with a high-order FD method in terms of the required number
of points per wavelength.

6.2 Mesh design

In a SEM, as in any method based upon a mesh of elements, the
design of the mesh is a critical and di¤cult step. The use of
hexahedral elements required in a SEMöno tetrahedra are
allowedöcomplicates matters further. Some expertise in mesh
Figure 7. Non-structured brick used to de¢ne a mesh with smaller
design or pre-processing of the model is required to ensure that
elements at the top of the structure, where velocities are usually
the resolution provided by the mesh, in particular the number smaller, and with bigger elements at the bottom, where velocities are
of gridpoints per wavelength, will be su¤cient for the problem. usually larger. We apply a geometrical grid doubling ¢rst in one of the
For example, densi¢cation of the mesh is usually needed near horizontal directions and then in the other. Here this coarsening in
the free surface, in areas of slow wave speed or strong contrasts depth has been applied twice; therefore, the elements at the bottom are
in wave speed, in regions of steep topography, or near faults. In four times bigger in the horizontal direction than the surface elements.

ß 1999 RAS, GJI 139, 806^822

814 D. Komatitsch and J. Tromp

is used in a FEM. However, even on a square with homo-

7.1 Layer-cake models
geneous elastic properties, the Gauss^Lobatto^Legendre
rule is exact only for polynomials of degree 2n` {1. Each To demonstrate that the non-structured mesh described in
integration on the reference element involves the product of Section 6.2 and represented in Fig. 7 is e¤cient for typical
two polynomials of degree n` : the displacement and the test layered geological structures, we ¢rst study a simple model con-
function. The integration of the resulting polynomial of degree sisting of a layer over a half-space, as shown in Fig. 8 (left). The
2n` is never exact, even in this simple case. For deformed horizontal size of the block used is 134 km|134 km, and the
elements there are additional errors related to curvature block extends to a depth of 60 km. Absorbing conditions are
(Maday & RÖnquist 1990); the same is true for elements with used on all sides of the model except the free surface. The non-
heterogeneous material properties. Thus, a diagonal mass structured mesh, whose coarsening in depth is implemented
matrix has been obtained by a process of subintegration. In this as shown in Fig. 7, is composed of 68 208 elements, using a
respect, the SEM is related to FEMs in which mass lumping is polynomial degree N~5, which results in a global grid com-

Downloaded from https://academic.oup.com/gji/article/139/3/806/587264 by Indian Institute of Technology Madras user on 08 November 2023
used to avoid the costly resolution of the non-diagonal system posed of 8 743 801 points. In the ¢rst simulation, the source is
resulting from the use of Gauss quadrature (Cohen et al. 1993). a vertical force located in the half-space, in the middle of the
It is interesting to note that even though a SEM exhibits high grid at a depth of 25.05 km. Therefore, the solution will include
spatial accuracy, the time marching scheme we use to solve the strong multiples in addition to the direct P and S waves. The
resulting ordinary di¡erential equation in time (32) is a simple time variation of the source is a Ricker wavelet, that is, the
second-order FD scheme. Therefore, in the long run the overall second derivative of a Gaussian, with a dominant frequency
accuracy of the simulation is often mainly governed by the time f0 ~0:4 Hz, and therefore a maximum frequency of the order
scheme, and in this respect the use of more accurate, higher- of 1 Hz. The onset time of the source is t0 ~2:6 s. The time step
order schemes, as suggested for instance by Tarnow & Simo used is *t~6:5 ms, and we propagate the signal for 6150 time
(1994), could be of interest. steps (i.e. 40 s). A line of receivers is placed at the surface along
the y-axis at x~xmax /2~67 km. The receivers record the three
components of displacement.
7 N U M E R ICA L R E S U LT S
Traces recorded at two receivers along the receiver line are
In order to validate the method, we consider ¢ve numerical shown in Fig. 9 for two of the components of the displacement
tests. In the ¢rst two tests, we study the response of two vector, the third (tangential) component being zero by sym-
1-D layer-cake models for both a shallow and a deep source. metry. The ¢rst receiver is located at a horizontal distance of
Solutions based upon a combined discrete wavenumber/ 2.39 km from the source, the second at 31.11 km. The strong
re£ectivity method are used for comparison. In the third test direct P and S waves can be clearly observed, as well as
we study a dip-slip source in a homogeneous half-space in strong multiples generated by the layer. To check the accuracy
order to check the accuracy of the moment tensor source of the simulation, we compare the SEM results to those based
representation. Results based upon frequency^wavenumber upon a discrete wavenumber method (Bouchon 1981) used
and FD techniques are available for comparison. In the fourth in conjunction with a re£ectivity method (Mu«ller 1985). The
test, we consider a hemispherical crater embedded in a homo- agreement between the two results is very good, both for the
geneous half-space in order to demonstrate the accuracy of the direct waves and for the numerous multiples. Small parasitic
free-surface boundary condition in the presence of very steep phases re£ected from the approximate absorbing conditions
topography. Results based upon an approximate boundary in our simulation probably explain the small discrepancies
method are used as a reference. These ¢rst four tests are for observed in the later arrivals between t~30 and t~35 s.
purely elastic media. In the ¢fth and ¢nal test, we study strong For the same model, we now place the source inside the
attenuation in a 2-D homogeneous medium and compare the layer, close to the surface, at a depth of 536.1 m. Thus the
results to the analytical solution of the problem. response includes a very signi¢cant surface wave contribution,

Figure 8. 3-D models with 1-D velocity structure used to assess the e¤ciency of the non-structured brick of Fig. 7. We study a model consisting of a
layer over a half-space (left) as well as a three-layer model over a half-space (right). In both cases the horizontal size of the block is 134 km|134 km,
and it extends to a depth of 60 km.

ß 1999 RAS, GJI 139, 806^822

 Seismic spectral element method 815

Downloaded from https://academic.oup.com/gji/article/139/3/806/587264 by Indian Institute of Technology Madras user on 08 November 2023
Figure 9. Traces recorded at the surface along the y-axis at x~67 km for the model composed of a layer over a half-space (Fig. 8, left). The source is
a vertical force located in the middle of the block at a depth of 25.05 km. The two receivers are located at horizontal distances of 2.39 km (top) and
31.11 km (bottom) from the source. The vertical (left) and radial (right) components of displacement are compared to the solution computed using a
discrete wavenumber method used in conjunction with the re£ectivity method. The third (tangential) component is zero by symmetry. The direct
P wave can be mainly observed on the vertical component, while the direct S wave has signi¢cant amplitude on both. Numerous strong multiples are
clearly visible.

whose accuracy will again be checked against the discrete We compute a very precise reference solution using a 1-D FD
wavenumber/re£ectivity method. Apart from the depth of the code with several hundred points per minimum wavelength.
source, all parameters remain the same as in the ¢rst simu- Fig. 11 shows the comparison at the surface for an incident
lation. Traces recorded at the surface at two receivers located, plane P wave and for an incident plane S wave. Numerous
respectively, at horizontal distances of 2.39 and 23.93 km from multiples can be observed after the ¢rst arrival. The agreement
the source are shown in Fig. 10. The main phase is a strong with the FD solution is excellent. The ampli¢cation at the
surface wave, whose amplitude is much larger than that of the surface is very signi¢cant: the maximum displacement recorded
direct P and S waves. Signi¢cant surface wave dispersion can is roughly 3.8 in both cases, for an incident value of 1.
be observed due to the presence of the low-velocity layer over We use the same three-layer model with a vertical force in
the faster half-space. The SEM solution is again in very good the middle of the block at a depth of 20.16 km. Traces recorded
agreement with the discrete wavenumber/re£ectivity reference. at two receivers, located, respectively, at horizontal distances
In a second simulation, we study the more complex model of 2.39 and 31.11 km from the source, are shown in Fig. 12,
consisting of three layers over a half-space illustrated in Fig. 8 as well as a comparison with results obtained based upon
(right). The non-structured mesh, which is based on the brick the discrete wavenumber/re£ectivity method. The agreement is
represented in Fig. 7, honours the three interfaces of the model again very good, even for this di¤cult structure, except for
at depths of 3, 12 and 25 km respectively. The grid doubling some small artefacts re£ected o¡ the absorbing boundaries in
is applied twice in the layer situated between 3 and 12 km. the SEM results between t~30 and t~35 s.
Apart from the number of layers and their properties, all other The above tests demonstrate that a SEM is capable of
parameters remain the same as before. We ¢rst study the accurately simulating wave propagation in 3-D models at a
displacement recorded at the surface in the case of vertically reasonable cost. We implemented the parallel algorithm based
incident plane P or S waves coming from the bottom of the upon OpenMP directives on shared memory architectures and
model. This test is interesting because, since the velocity model the Message-Passing Interface (MPI) on distributed memory
is 1-D and the incidence is vertical, no mode conversions occur, architectures. The total CPU time of each of these simulations
and therefore the solution is a superposition of pure P or using the MPI implementation on an eight-node Dec Alpha
pure S waves, which allows us to validate them separately. was roughly 8 hr. We obtained a total sustained performance
Furthermore, for normal incidence the absorbing condition of 1.3 Giga£ops, a parallel speedup of 7.3, and a parallel
(11) used at the bottom of the grid is exact, therefore in this e¤ciency of 91 per cent. The total memory needed was roughly
test we have no spurious re£ections o¡ the boundaries. On the 1 Gbytes. The MPI code was also successfully run on a
vertical edges of the grid we use periodic boundary conditions. network of PCs under Linux (a so-called Beowulf machine).

ß 1999 RAS, GJI 139, 806^822

816 D. Komatitsch and J. Tromp

Downloaded from https://academic.oup.com/gji/article/139/3/806/587264 by Indian Institute of Technology Madras user on 08 November 2023
Figure 10. Traces recorded at the surface along the y-axis at x~67 km for the model composed of a layer over a half-space (Fig. 8, left). The source
is a vertical force located in the sedimentary layer in the middle of the block at a depth of 536.1 m. The two receivers are located at horizontal
distances of 2.39 km (top) and 23.93 km (bottom) from the source. The main event is a strong dispersive surface wave whose amplitude is several times
bigger than that of the direct P and S waves. The vertical (left) and radial (right) components of displacement are compared to the solution computed
using a discrete wavenumber method used in conjunction with the re£ectivity method. The third (tangential) component is zero by symmetry.

frequency^wavenumber (FK) technique; his results are used as

7.2 Moment tensor source
a reference. The SEM mesh is a structured block of 42|42|22
As discussed in the Appendix, a moment tensor source can elements, with a polynomial degree N~5 used in each element.
be naturally introduced in a SEM. To check the accuracy of Therefore, the global grid is composed of 4 941 831 gridpoints.
the approach, we use a homogeneous half-space with P-wave The mesh represents a block of size 42 km|42 km|22 km. The
velocity cp ~4000 m s{1, S-wave velocity cs ~2300 m s{1 and time step used is *t~5 ms, and we propagate the signal for
density o~1800 kg m{3. The source is placed at a depth 2400 time steps (i.e. 12 s). The time dependence of the velocity
of 2.5 km, and receivers are placed exactly at the surface, source is a triangle with base width tb ~1 s and onset time
at a horizontal distance of 10 km. We use a pure dip-slip t0 ~0:55 s. Results are low-pass ¢ltered at a cut-o¡ frequency
source (strike 900, dip 900, rake 900) with scalar moment of f0 ~1 Hz using a Butterworth ¢lter, as in Graves (1996). In
M0 ~1016 N m, and record the vertical and radial components Fig. 13, we show the components of velocity obtained based
of velocity at an azimuth of 00 and the tangential component upon the three techniques, without any normalization of the
at an azimuth of 900. This problem has been studied pre- amplitude. The overall agreement is good, which validates the
viously by Graves (1996) using both an FD technique and a implementation of the moment tensor source in the SEM.

Figure 11. Response recorded at the surface of the model composed of three layers over a half-space (Fig. 8, right) due to plane P (left) and S (right)
waves incident perpendicularly from below. A highly accurate 1-D FD solution is shown for comparison. The absorbing bottom boundary condition
is exact in this case; periodic boundary conditions are applied at the four sides of the block. The incident plane wave has an amplitude of 1 and is
ampli¢ed by a factor of roughly 3.8 near the surface.

ß 1999 RAS, GJI 139, 806^822

 Seismic spectral element method 817

Downloaded from https://academic.oup.com/gji/article/139/3/806/587264 by Indian Institute of Technology Madras user on 08 November 2023
Figure 12. Traces recorded at the surface along the y-axis at x~67 km for the model composed of three layers over a half-space (Fig. 8, right). The
source is a vertical force located in the middle of the block at a depth of 20.16 km. The two receivers are located at horizontal distances of 2.39 km
(top) and 31.11 km (bottom) from the source. The vertical (left) and radial (right) components of displacement are compared to the solution computed
using a discrete wavenumber method in conjunction with the re£ectivity method. The third (tangential) component is zero by symmetry.

Some non-causal oscillations are present in the FK reference, mode conversions correctly. We compute the amplitude of the
in particular on the tangential component, which may explain displacement at the surface along a pro¢le for two values of
the small discrepancies observed in the amplitude. These the normalized frequency, g~0:25 and g~0:50, as a function
oscillations could perhaps be reduced by using a discrete wave- of the normalized horizontal coordinate x/a between 0 and 2,
number method to compute the reference. The FD results are starting at the centre of the crater. Poisson's ratio is equal to
slightly less accurate, mainly for the radial component, and show 0.25. The wavelengths considered are of the order of the radius
some parasitic later arrivals around t~8 s that are probably of the crater. Therefore, the main concern in this simulation
due to spurious re£ections o¡ the absorbing boundaries. is not the resolution of the computations, that is, number
of points per wavelength, but the correct description of the
geometry of the hemisphere. In particular, this has a signi¢-
7.3 Hemispherical crater
cant in£uence on the correct modelling of the surface waves
Sänchez-Sesma (1983) studied the response of a hemispherical generated by the incident P wave, travelling back and forth
crater in a homogeneous half-space to a vertically incident along the crater. We can therefore use very big elements in the
plane P wave based upon an approximate boundary method. half-space, but need to use small enough curved elements close
He presented the amplitude of the displacement recorded at to the crater to describe its shape correctly. To avoid artefacts
the surface of the crater for di¡erent normalized frequencies related to arti¢cial boundaries, considering a crater with a
g~2a/jP , where a denotes the radius of the crater and jP the radius of a~1 km, we extend the model to a horizontal size of
wavelength of the incident P wave. His results have been used 32 km|32 km, and a depth of 16 km. The mesh is composed
as a reference in several subsequent studies (Mossessian & of 1800 elements, with a polynomial degree N~4 in each
Dravinski 1989; Luzön et al. 1997). element; the global mesh contains 120 089 gridpoints. Such
Mesh generation based upon conforming hexahedra is non- a small value is again due to the fact that we have used a
trivial for a sphere or a hemisphere. Fortunately, an elegant small number of big elements in this simulation. Periodic con-
analytical mesh called the `cubed sphere' has recently been ditions are used on the four vertical sides of the global grid.
developed (Ronchi et al. 1996). Such a mesh was ¢rst used Considering a P-wave velocity of cp ~1732 m s{1 and an S-wave
by Chaljub & Vilotte (1998) for global wave propagation velocity of cs ~1000 m s{1 , the time step used is *t~5 ms, and
problems. We use the ¢ve sides of half a cubed sphere to mesh the signal is propagated for 3200 time steps (i.e. 16 s). The
the hemisphere, as shown in Fig. 14. The global mesh is there- density is 1000 kg m{3 . The source is an incident plane P wave,
fore composed of ¢ve structured blocks. A desirable property whichpis
 a Ricker wavelet in time with dominant frequency
of such a mesh is that it takes into account the vertical edges of f0 ~ 3/4 Hz. It is introduced in the SEM as initial displace-
the crater at the free surface and is naturally re¢ned in the ment and velocity ¢elds in the time scheme. Seismograms are
vicinity of the crater, where a high degree of accuracy is needed then converted to the frequency domain for comparison with
to describe the geometry and to resolve ampli¢cation and Sänchez-Sesma (1983).

ß 1999 RAS, GJI 139, 806^822

818 D. Komatitsch and J. Tromp

Downloaded from https://academic.oup.com/gji/article/139/3/806/587264 by Indian Institute of Technology Madras user on 08 November 2023
Figure 14. Close-up of the non-structured mesh based upon half a
`cubed sphere' used to de¢ne the grid for a hemispherical crater in
a homogeneous half-space. The mesh is composed of ¢ve structured
blocks. It is naturally adapted to the sharp edges of the crater and is
densi¢ed in its vicinity. One can see the very big aspect ratio of the
elements close to the crater, because in this simulation the main con-
cern is the correct description of the geometry of the crater, not the
number of points per wavelength, due to the low frequencies con-
sidered. For display purposes, one of the structured blocks has been
removed.

sented by a constant QP ^30 and a constant QS ^20 is

Figure 13. Spectral element (solid line), FK (dashed line) and FD
introduced. The relaxed (elastic) velocities of the medium are
(dotted line) results obtained at the surface of a homogeneous half-
space in the case of a pure dip-slip moment tensor source situated at a
cp ~3000 m s{1 for the P wave and cs ~2000 m s{1 for the
depth of 2.5 km. The receivers are located at a horizontal distance of S wave. The density is 2000 kg m{3 . In light of the strong
10 km. The vertical (top) and radial (middle) components of velocity are attenuation, we expect very signi¢cant physical velocity dis-
recorded at an azimuth of 00, while the tangential component (bottom) persion. Carcione et al. (1988) studied this problem with
is recorded at an azimuth of 900. The source is a triangular velocity a Fourier method for a point force source, and also derived
pulse in time, and the results are low-pass ¢ltered at a cut-o¡ frequency the analytical solution using the correspondence principle to
of 1 Hz. The FK and FD results are taken from Graves (1996). compute the Green's function in the frequency domain. Their
analytical results are used here for comparison. The source is a
vertical force placed exactly in the middle of the model, at
Fig. 15 shows a comparison between the two methods for xs ~zs ~1000 m. The time variation of the source is a Ricker
g~0:25 and g~0:50. The agreement is excellent. In particular, wavelet with dominant frequency f0 ~18 Hz and onset time
the strong ampli¢cation close to the edges of the crater t0 ~0.06 s. The source we use has the same frequency content
is reproduced well. Note that the ampli¢cation level of the as that used by Carcione et al. (1988), that is, it has signi¢cant
vertical component reaches a very high value (^3.2) in energy up to roughly 50 Hz. We do not use exactly the same
the centre for g~0:50. The horizontal component is zero in the source in order to avoid energy at zero frequency, and thus
centre by symmetry. Close to the edge, the horizontal com- circumvent a potential problem in the computation of the
ponent reaches 60 per cent of the amplitude of the incident analytical solution in the frequency domain (Carcione et al.
wave for g~0:50. The small discrepancy that can be observed 1988, Appendix B). The constant values QP ^30 and QS ^20
between x/a~1 and x/a~1:5 km for g~0:50 may be due to an are mimicked using two standard linear solids; we use the
underestimation in Sänchez-Sesma (1983) because subsequent optimal relaxation times computed by Carcione et al. (1988)
studies of Mossessian & Dravinski (1989) and Luzön et al. for these two mechanisms.
(1997) also predict higher values in this region. The medium is discretized using 44|44 spectral elements,
with a polynomial degree N~5. Therefore, the global grid
comprises 221|221~48 841 points. As mentioned previously,
7.4 Homogeneous model with strong attenuation
we use an explicit second-order FD time scheme to march the
In order to validate the way in which we have introduced weak form of the momentum equation, and a fourth-order
attenuation in the SEM, we consider a 2-D homogeneous Runge^Kutta scheme to march the strong form of the memory
medium of size 2000 m|2000 m. Strong attenuation repre- variable equations. Both schemes use the same time step

ß 1999 RAS, GJI 139, 806^822

 Seismic spectral element method 819

Downloaded from https://academic.oup.com/gji/article/139/3/806/587264 by Indian Institute of Technology Madras user on 08 November 2023
Figure 16. Amplitude of the horizontal (top) and vertical (bottom)
components of displacement recorded in a 2-D homogeneous medium
of size 2000 m| 2000 m with constant QP ^30 and QS ^20. The
Figure 15. Amplitudes of the two components of displacement source is a vertical force placed exactly in the middle of the model.
recorded along the crater, from x/a~0 (centre of the crater) to The constant Q values are mimicked using two standard linear solids.
x/a~2 km. The vertical and radial components are displayed. The We present both the spectral element solution (solid line) and the
third (tangential) component is zero by symmetry. The results are analytical solution derived by Carcione et al. (1988) (dashed line) for a
shown for two normalized frequencies, g~0:25 (top) and g~0:50 receiver located at xr ~zr ~1500 m, that is, at a distance of 707.1 m
(bottom). The symbols are used to show the SEM results. The solid and from the source. The very strong e¡ect of attenuation can be observed
dashed lines are the results of Sänchez-Sesma (1983) obtained based by comparing these results to the SEM solution computed for an elastic
upon an approximate boundary method. medium with the same relaxed material properties (dotted line).

*t~0:75 ms. We propagate the signal for 0.75 s, i.e. 1000 time velocity, the method accurately models the propagation of both
steps. In Fig. 16 we present both the SEM solution and the body and surface waves for shallow as well as deep sources.
analytical solution for a receiver located at xr ~zr ~1500 m, The technique accommodates point force and moment tensor
that is, at a distance of 707.1 m from the source. The agreement sources, and can easily be extended to include ¢nite sources.
is very good, which validates the approach used to incorporate The accuracy of the free-surface implementation was demon-
attenuation. To show how strong the attenuation and associated strated for the problem of a hemispherical crater embedded in a
velocity dispersion are, we also plot the SEM solution com- homogeneous half-space, for which an approximate boundary
puted for an elastic medium with the same relaxed material method was used as a reference. The e¡ects of attenuation were
properties (that is, the medium obtained when Q??). One can incorporated based upon an absorption-band model. For a
see, for instance, that the amplitude of the S wave is reduced by highly attenuating model with associated strong dispersion the
a factor of more than two. accuracy of the method was demonstrated by comparison with
an analytical solution.
In order to model wave propagation in realistic 3-D earth
8 C O NCLUS IO N S
models, for instance to determine seismic risk associated
We have presented a detailed introduction to the spectral with potential earthquakes in metropolitan areas such as Los
element method for 3-D seismic wave propagation. The method Angeles, Tokyo or Mexico City, several di¤cult problems need
incorporates surface topography, attenuation and anisotropy, to be addressed. First, we need to show that the SEM is capable
and accurately represents surface waves. We have bench- of dealing with highly heterogeneous 3-D structures, including
marked the method against a discrete wavenumber/re£ectivity regions with high Poisson's ratio. This problem has been
method for layer-cake models. Even in the presence of very sharp partially addressed in Komatitsch (1997) and Komatitsch &
discontinuities, for example, a factor of 3 increase in P-wave Vilotte (1998), but it remains to be shown that the method can

ß 1999 RAS, GJI 139, 806^822

820 D. Komatitsch and J. Tromp

also deal with more realistic models. Second, detailed 3-D Clayton, R. & Engquist, B., 1977. Absorbing boundary conditions
models of such basins, constructed based upon geological, for acoustic and elastic wave equations, Bull. seism. Soc. Am., 67,
well-log and seismic re£ection and refraction data, need to be 1529^1540.
Cohen, G., Joly, P. & Tordjman, N., 1993. Construction and analysis
constructed, and for larger earthquakes the details of the
of higher-order ¢nite elements with mass lumping for the wave
rupture also have to be determined; several studies have been
equation, in 2nd Int. Conf. on Mathematical and Numerical Aspects
dedicated to these problems in recent years (e.g. Wald 1996; of Wave Propagation, ed. Kleinman, R., pp. 152^160, SIAM,
Olsen & Archuleta 1996; Pitarka et al. 1998; Wald & Graves Philadelphia.
1998). Day, S.M., 1998. E¤cient simulation of constant Q using coarse-
As underlined in the innovative work of Chaljub & Vilotte grained memory variables, Bull. seism. Soc. Am., 88, 1051^1062.
(1998), in the context of global seismology several obstacles need Dhatt, G. & Touzot, G., 1984. The Finite Element Method Displayed,
to be overcome. At long periods, self-gravitation is important John Wiley, New York.
and needs to be accommodated in the weak formulation of the Emmerich, H. & Korn, M., 1987. Incorporation of attenuation into

Downloaded from https://academic.oup.com/gji/article/139/3/806/587264 by Indian Institute of Technology Madras user on 08 November 2023
equations of motion. To maintain a relatively constant number time-domain computations of seismic wave ¢elds, Geophysics, 52,
of gridpoints per wavelength, the mesh has to be coarsened 1252^1264.
Faccioli, E., Maggio, F., Paolucci, R. & Quarteroni, A., 1997. 2D
several times as a function of depth. Finally, the core^mantle
and 3D elastic wave propagation by a pseudo-spectral domain
boundary is an extremely sharp £uid^solid interface that decomposition method, J. Seism., 1, 237^251.
needs to be incorporated accurately. Even though the spectral Fischer, P.F. & RÖnquist, E.M., 1994. Spectral element methods for
element method has been successfully applied for problems large scale parallel Navier-Stokes calculations, Comp. Meth. appl.
involving a homogeneous £uid (Komatitsch et al. 1999a), it has Mech. Eng., 116, 69^76.
to be extended to include inhomogeneous £uids. Furumura, T., Kennett, B.L.N. & Furumura, M., 1998. Seismic wave-
¢eld calculation for laterally heterogeneous whole earth models
using the pseudospectral method, Geophys. J. Int., 135, 845^860.
AC K NOW L E D GM E N T S Geller, R.J. & Ohminato, T., 1994. Computation of synthetic seismo-
We are very grateful to Michel Bouchon for numerous dis- grams and their partial derivatives for heterogeneous media with
cussions and for providing us with his 3-D DWN code. We arbitrary natural boundary conditions using the direct solution
method, Geophys. J. Int., 116, 421^446.
would also like to thank Emmanuel Chaljub and Jean-Pierre
Graves, R.W., 1996. Simulating seismic wave propagation in 3D elastic
Vilotte for fruitful discussion, in particular regarding the use of
media using staggered-grid ¢nite di¡erences, Bull. seism. Soc. Am.,
the cubed sphere, Robert W. Graves for providing us with his 86, 1091^1106.
results for the moment tensor source, and Paul F. Fischer for Guo, B. & Babus̄ka, I., 1986. The h-p version of the ¢nite element
providing us with his non-structured global numbering code. method, Comp. Mech., 1, 21^41.
Discussions with Francisco-Josë Sänchez-Sesma regarding the Hughes, T.J.R., 1987. The Finite Element Method, Linear Static and
solution of the hemispherical crater problem, and with Josë Dynamic Finite Element Analysis, Prentice-Hall, Englewood Cli¡s,
M. Carcione and Marwan Charara regarding the analytical NJ.
solution of the attenuation problem are also acknowledged. Igel, H., Mora, P. & Riollet, B., 1995. Anisotropic wave propagation
We thank Robert W. Graves and an anonymous reviewer through ¢nite-di¡erence grids, Geophysics, 60, 1203^1216.
Kay, I. & Krebes, E.S., 1999. Applying ¢nite element analysis to the
for constructive comments that improved the manuscript.
memory variable formulation of wave propagation in anelastic
The Caltech computer center CACR kindly gave us access to
media, Geophysics, 64, 300^307.
their PC cluster, and the DMPN at IPGP provided access Kelly, K.R., Ward, R.W., Treitel, S. & Alford, R.M., 1976. Synthetic
to their DEC Alpha. This work was funded in part by the seismograms: a ¢nite di¡erence approach, Geophysics, 41, 2^27.
David and Lucile Packard Foundation, NSF and NEHRP. Komatitsch, D., 1997. Mëthodes spectrales et ëlëments spectraux pour
l'ëquation de l'ëlastodynamique 2D et 3D en milieu hëtëroge©ne,
PhD thesis, Institut de Physique du Globe de Paris, Paris.
R E F E R E NCE S
Komatitsch, D. & Vilotte, J.P., 1998. The spectral element method:
Bao, H., Bielak, J., Ghattas, O., Kallivokas, L.F., O'Hallaron, D.R., an e¤cient tool to simulate the seismic response of 2D and 3D
Shewchuk, J.R. & Xu, J., 1998. Large-scale simulation of elastic geological structures, Bull. seism. Soc. Am., 88, 368^392.
wave propagation in heterogeneous media on parallel computers, Komatitsch, D., Coutel, F. & Mora, P., 1996. Tensorial formulation of
Comp. Meth. appl. Mech. Eng., 152, 85^102. the wave equation for modelling curved interfaces, Geophys. J. Int.,
Bernardi, C., Debit, N. & Maday, Y., 1990. Coupling ¢nite element and 127, 156^168.
spectral methods: ¢rst results, Math. Comput., 54, 21^39. Komatitsch, D., Barnes, C. & Tromp, J. 1999a. Wave propagation near
Bouchon, M., 1981. A simple method to calculate Green's functions for a £uid-solid interface: a spectral element approach, Geophysics,
elastic layered media, Bull. seism. Soc. Am., 71, 959^971. in press.
Bouchon, M., Schultz, C.A. & To«ksoz, M.N., 1996. E¡ect of three- Komatitsch, D., Barnes, C. & Tromp, J. 1999b. Simulation of aniso-
dimensional topography on seismic motion, J. geophys. Res., 101, tropic wave propagation based upon a spectral element method,
5835^5846. Geophysics, submitted.
Canuto, C., Hussaini, M.Y., Quarteroni, A. & Zang, T.A., 1988. Komatitsch, D., Vilotte, J.P., Vai, R., Castillo-Covarrubias, J.M. &
Spectral Methods in Fluid Dynamics, Springer-Verlag, New York. Sänchez-Sesma, F.J., 1999c. The spectral element method for elastic
Carcione, J.M., 1994. The wave equation in generalized coordinates, wave equations: application to 2D and 3D seismic problems, Int. J.
Geophysics, 59, 1911^1919. Num. Meth. Eng., 45, 1139^1164.
Carcione, J.M., Koslo¡, D. & Koslo¡, R., 1988. Wave propagation Lahaye, D.J.P., Maggio, F. & Quarteroni, A., 1997. Hybrid ¢nite
simulation in a linear viscoelastic medium, Geophys. J. Int., 95, element-spectral element approximation of wave propagation
597^611. problems, East-West J. Num. Math., 5, 265^289.
Chaljub, E. & Vilotte, J.P., 1998. 3D wave propagation in a spherical Liu, H.P., Anderson, D.L. & Kanamori, H., 1976. Velocity dispersion
earth model using the spectral element method, EOS, Trans. Am. due to anelasticity: implications for seismology and mantle com-
geophys. Un., 79, F625^F626. position, Geophys. J. R. astr. Soc., 47, 41^58.

ß 1999 RAS, GJI 139, 806^822

 Seismic spectral element method 821

Luzön, F., Sänchez-Sesma, F.J., Rodr|¨ guez-Zü·iga, J.L., Posadas, A.M., Tessmer, E. & Koslo¡, D., 1994. 3-D elastic modeling with surface
Garc|¨ a, J.M., Mart|¨ n, J., Romacho, M.D. & Navarro, M., topography by a Chebyshev spectral method, Geophysics, 59,
1997. Di¡raction of P, S and Rayleigh waves by three-dimensional 464^473.
topographies, Geophys. J. Int., 129, 571^578. Tessmer, E., Kessler, D., Koslo¡, D. & Behle, A., 1992. Multi-domain
Maday, Y. & RÖnquist, E.M., 1990. Optimal error analysis of spectral Chebyshev-Fourier method for the solution of the equations of
methods with emphasis on non-constant coe¤cients and deformed motion of dynamic elasticity, J. Comput. Phys., 100, 355^363.
geometries, Comp. Meth. appl. Mech. Eng., 80, 91^115. Virieux, J., 1986. P-SV wave propagation in heterogeneous media:
Marfurt, K.J., 1984. Accuracy of ¢nite-di¡erence and ¢nite-element velocity-stress ¢nite-di¡erence method, Geophysics, 51, 889^901.
modeling of the scalar wave equation, Geophysics, 49, 533^549. Wald, D.J., 1996. Slip history of the 1995 Kobe, Japan, earthquake
Moczo, P., Bystrickÿ, E., Kristek, J., Carcione, J.M. & Bouchon, M., determined from strong motion, teleseismic, and geodetic data,
1997. Hybrid modeling of P-SV seismic motion at inhomogeneous J. Phys. Earth, 44, 489^503.
viscoelastic topographic structures, Bull. seism. Soc. Am., 87, Wald, D.J. & Graves, R.W., 1998. The seismic responses of the Los
1305^1323. Angeles basin, California, Bull. seism. Soc. Am., 88, 337^356.

Downloaded from https://academic.oup.com/gji/article/139/3/806/587264 by Indian Institute of Technology Madras user on 08 November 2023
Mossessian, T.K. & Dravinski, M., 1989. Scattering of elastic waves by Zeng, X., 1996. Finite di¡erence modeling of viscoelastic wave
three-dimensional surface topographies, Wave Motion, 11, 579^592. propagation in a generally heterogeneous medium in the time
Mu«ller, G., 1985. The re£ectivity method: a tutorial, J. Geophys., 58, domain, and a dissection method in the frequency domain, PhD
153^174. thesis, University of Toronto, Canada.
Ohminato, T. & Chouet, B.A., 1997. A free-surface boundary condition Zienkiewicz, O.C., 1977. The Finite Element Method in Engineering
for including 3D topography in the ¢nite di¡erence method, Bull. Science, 3rd edn, McGraw-Hill, New York.
seism. Soc. Am., 87, 494^515.
Olsen, K.B. & Archuleta, R.J., 1996. 3-D simulation of earthquakes on
the Los Angeles fault system, Bull. seism. Soc. Am., 86, 575^596.
A PPE N D I X A : D ETA I L E D D E SC R IP T I O N
Paolucci, R., Faccioli, E. & Maggio, F., 1999. 3D response analysis of
OF T H E W E A K FORMU L AT I O N
an instrumented hill at Matsuzaki, Japan, by a spectral method,
J. Seism., 3, 191^209. In this Appendix we present detailed expressions for each
Patera, A.T., 1984. A spectral element method for £uid dynamics: of the terms in the weak formulation of the equations of
laminar £ow in a channel expansion, J. Comput. Phys., 54, 468^488. motion (12), suitable for numerical evaluation at the elemental
Pitarka, A., Irikura, K., Iwata, T. & Sekiguchi, H., 1998. Three-
level. The expression for the left-hand side of (12), the mass
dimensional simulation of the near-fault ground motion for the 1995
Hyogo-ken Nanbu (Kobe), Japan, earthquake, Bull. seism. Soc.
matrix, was discussed in Section 5. The remaining three terms
Am., 88, 428^440. on the right-hand side are presented here.
Priolo, E., Carcione, J.M. & Seriani, G., 1994. Numerical simulation The ¢rst integral that needs to be evaluated at the elemental
of interface waves by high-order spectral modeling techniques, level is
J. acoust. Soc. Am., 95, 681^693.
Quarteroni, A., Tagliani, A. & Zampieri, E., 1998. Generalized
Galerkin approximations of elastic waves with absorbing boundary =w5T d 3 x . (A1)
)e
conditions, Comp. Meth. appl. Mech. Eng., 163, 323^341.
Robertsson, J.O.A., 1996. A numerical free-surface condition for
This integral is often referred to as the elemental sti¡ness
elastic/viscoelastic ¢nite-di¡erence modeling in the presence of
topography, Geophysics, 61, 1921^1934. matrix. The ¢rst step is to calculate the nine elements of the
Robertsson, J.O.A., Blanch, J.O. & Symes, W.W., 1994. Viscoelastic displacement gradient =s on the element )e . This gives
¢nite-di¡erence modeling, Geophysics, 59, 1444^1456. " #
Ronchi, C., Ianoco, R. & Paolucci, P.S., 1996. The `Cubed Sphere': X
n`

a new method for the solution of partial di¡erential equations in Li sj (x(ma , gb , fc ), t)~ spbc 0
j (t)`p (ma ) Li m(ma , gb , fc )
spherical geometry, J. Comput. Phys., 124, 93^114. p~0
Sänchez-Sesma, F.J., 1983. Di¡raction of elastic waves by three- " #
dimensional surface irregularities, Bull. seism. Soc. Am., 73, X
n`

1621^1636. z sapc 0
j (t)`p (gb ) Li g(ma , gb , fc )
p~0
Seriani, G., 1997. A parallel spectral element method for acoustic wave
modeling, J. Comput. Acoust., 5, 53^69. " #
Seriani, G., 1998. 3-D large-scale wave propagation modeling by X
n`

spectral element method on Cray T3E multiprocessor, Comp. Meth.

z sabp 0
j (t)`p (fc ) Li f(ma , gb , fc ) . (A2)
p~0
appl. Mech. Eng., 164, 235^247.
Seriani, G. & Priolo, E., 1994. Spectral element method for acoustic
wave simulation in heterogeneous media, Finite Elements in This calculation requires knowledge of the nine elements of the
Analysis and Design, 16, 337^348. inverse Jacobian matrix Lî/Lx. Next, one calculates the six
Seriani, G., Priolo, E., Carcione, J.M. & Padovani, E., 1992. High- elements of the symmetric stress tensor T on the element:
order spectral element method for elastic wave modeling, in
Expanded abstracts of the SEG, pp. 1285^1288, 62nd Int. Mtng of T(x(ma , gb , fc ), t)~c(x(ma , gb , fc ))5=s(x(ma , gb , fc ), t) . (A3)
the SEG, New-Orleans.
Seriani, G., Priolo, E. & Pregarz, A., 1995. Modelling waves in aniso-
This requires knowledge of the previously calculated dis-
tropic media by a spectral element method, in Third Int. Conf. on
placement gradient (A2) and of the elastic tensor c at the
Mathematical and Numerical Aspects of Wave Propagation, ed.
Cohen, G., pp. 289^298, SIAM, Philadelphia. Gauss^Lobatto^Legendre integration points. The formulation
Tarnow, N. & Simo, J.C., 1994. How to render second-order accurate is not limited to isotropic media, or to anisotropic media with
time-stepping algorithms fourth-order accurate while retaining the a high degree of symmetry, as is often the case for other
stability and conservation properties, Comp. Meth. appl. Mech. numerical methods. The earth model may be fully hetero-
Eng., 115, 233^252. geneous, that is, c need not be constant inside an element. The

ß 1999 RAS, GJI 139, 806^822

822 D. Komatitsch and J. Tromp

integrand =w5T may be written in the form reduces to

!
X 3 X 3 X3
Lwi X
3
a bs cs 0
=w5T~ Tij Lj wi ~ Tij Lj mk M5=w(xs )& wabc
i [Gi1
s
`a (mas )dbbs dccs
i, j~1 i,k~1 j~1
Lmk i~1

X a bs cs a bs cs
3
Lwi zGi2s daas `0b (gbs )dccs zGi3s daas dbbs `0c (fcs )] .
~ Fik , (A4)
i,k~1
Lmk
(A11)
where In the case of a ¢nite fault plane the mesh should be designed
X
3 such that the fault coincides with the side of an element, that
Fik ~ Tij Lj mk . (A5) is, the fault is represented by a ¢nite number of `boundary

Downloaded from https://academic.oup.com/gji/article/139/3/806/587264 by Indian Institute of Technology Madras user on 08 November 2023
j~1 elements'. In that case we have
The next step is to calculate the nine matrix elements X
3  X
n`
Fik on the Gauss^Lobatto^Legendre integration points: m(xs , t)5=w(xs ) d 2 xs & wabc
i ub uas Jbas b gai1s bc `0a (mas )
Ss
Fikpql ~Fik (x(mp , gq , fl )); this requires knowledge of the stress i~1 as ~0

tensor T computed in (A3) and of the inverse Jacobian X

n`
ab ab c
matrix Lî/Lx. The integral (A1) may now be rewritten in the zua ubs Jb s gi2 s `0b (gbs )
form bs ~0

X
3 X 3 1 1 1 
Lwi 3 Lwi abc
Fik d x~ Fik Je dm dg df . zua ub Jbab gi3 s `0c (fcs ) , (A12)
i,k~1 )e Lmk i,k~1 {1 {1 {1 Lmk

(A6) where

Upon substituting the test vector (35) in (A6) and introducing X

3
gik ~ mij Lj mk (A13)
the Gauss^Lobatto^Legendre integration rule, we ¢nd that j~1

X3  X
n`
0 and mij denotes the elements of the moment density tensor.
wabc ua0 Jea bc Fi1a bc `0a (ma0 )
0
=w5T d 3 x& i ub uc
)e Gridpoints within the fault plane are denoted by mas and gbs ,
i~1 a0 ~0
and gridpoints perpendicular to the fault plane are denoted
X
n`
0 0 by fcs .
zua uc ub0 Jeab c Fi2ab c `0b (gb0 )
Finally, the absorbing boundary integral in (12) may be
b0 ~0
approximated at the elemental level as
X
n` 
0
Jeabc Fi3abc `0c (fc0 )
0
zua ub u . (A7)
c0
o[on (nª . Lt s)nª zo1 ( tê 1 . Lt s) tê 1 zo2 ( tê 2 . Lt s) tê 2 ] . w d 2 x
c0 ~0
!e

The source term M5=w may be rewritten as "

X
3 X
3
! & wabc ua ub oabc oabc nª j s_abc
X 3 X 3 X3 i n nª i j
Lwi i~1 j~1
M5=w~ Mij Lj wi ~ Mij Lj mk
i, j~1 i,k~1 j~1
Lmk ! #
X
3 X
3

X
3 zoabc
1 tê 1i tê 1j s_abc abc
j zo2 tê 2i tê 2j s_abc
j Jbab . (A14)
Lwi
~ Gik , (A8) j~1 j~1
i,k~1
Lmk
Here ma and gb denote gridpoints within the absorbing boundary,
where and fc denotes gridpoints perpendicular to the boundary. Note
X
3 that, like the mass matrix, the absorbing boundary term is
Gik ~ Mij Lj mk . (A9) diagonal.
j~1 In an anelastic medium, the sti¡ness matrix is still given
by (A1), except that the stress tensor (A3) needs to be replaced
Upon de¢ning, as per usual, Gpql
ik ~Gik (x(mp , gq , fl )) and using by
the test vector (35),
 X T(x(ma , gb , fc ), t)~c(x(ma , gb , fc ))5=s(x(ma , gb , fc ), t)
X3 n`
M5=w(xs )& wabc
i `p (mas )`q (gbs )`l (fcs )
X
L
i~1 p,q,l~0
{ R` (x(ma , gb , fc ), t) , (A15)
|[Gpql 0
i1 `a (mas )`b (gbs )`c (fcs )
`~1

in accordance with (5). The implication is that the six linearly

zGpql 0
i2 `a (mas )`b (gbs )`c (fbs ) independent components of the symmetric memory tensor R`
 need to be stored on the grid. For this reason, the memory
zGpql ` (m )` (g )`
i3 a as b bs c cs
0
(f )] , (A10) requirements for an anelastic simulation increase substantially
over those for a purely elastic simulation, and it may in this
where x(mas , gbs , fcs )~xs . If the source location xs coincides respect be of interest to spread the memory variables across an
with one of the Gauss^Lobatto^Legendre points, (A10) element to obtain the expected behaviour (Day 1998).

ß 1999 RAS, GJI 139, 806^822